fairseq/models/roberta/hub_interface.py

Survived

Mutant 3282

--- fairseq/models/roberta/hub_interface.py
+++ fairseq/models/roberta/hub_interface.py
@@ -29,7 +29,6 @@
         # this is useful for determining the device
         self.register_buffer('_float_tensor', torch.tensor([0], dtype=torch.float))
 
-    @property
     def device(self):
         return self._float_tensor.device
 

Mutant 3283

--- fairseq/models/roberta/hub_interface.py
+++ fairseq/models/roberta/hub_interface.py
@@ -33,7 +33,7 @@
     def device(self):
         return self._float_tensor.device
 
-    def encode(self, sentence: str, *addl_sentences, no_separator=False) -> torch.LongTensor:
+    def encode(self, sentence: str, *addl_sentences, no_separator=True) -> torch.LongTensor:
         """
         BPE-encode a sentence (or multiple sentences).
 

Mutant 3284

--- fairseq/models/roberta/hub_interface.py
+++ fairseq/models/roberta/hub_interface.py
@@ -74,7 +74,7 @@
             return sentences[0]
         return sentences
 
-    def extract_features(self, tokens: torch.LongTensor, return_all_hiddens: bool = False) -> torch.Tensor:
+    def extract_features(self, tokens: torch.LongTensor, return_all_hiddens: bool = True) -> torch.Tensor:
         if tokens.dim() == 1:
             tokens = tokens.unsqueeze(0)
         if tokens.size(-1) > self.model.max_positions():

Mutant 3285

--- fairseq/models/roberta/hub_interface.py
+++ fairseq/models/roberta/hub_interface.py
@@ -100,7 +100,7 @@
             name, num_classes=num_classes, embedding_size=embedding_size, **kwargs
         )
 
-    def predict(self, head: str, tokens: torch.LongTensor, return_logits: bool = False):
+    def predict(self, head: str, tokens: torch.LongTensor, return_logits: bool = True):
         features = self.extract_features(tokens.to(device=self.device))
         logits = self.model.classification_heads[head](features)
         if return_logits:

Mutant 3286

--- fairseq/models/roberta/hub_interface.py
+++ fairseq/models/roberta/hub_interface.py
@@ -107,7 +107,7 @@
             return logits
         return F.log_softmax(logits, dim=-1)
 
-    def extract_features_aligned_to_words(self, sentence: str, return_all_hiddens: bool = False) -> torch.Tensor:
+    def extract_features_aligned_to_words(self, sentence: str, return_all_hiddens: bool = True) -> torch.Tensor:
         """Extract RoBERTa features, aligned to spaCy's word-level tokenizer."""
         from fairseq.models.roberta import alignment_utils
         from spacy.tokens import Doc

Mutant 3287

--- fairseq/models/roberta/hub_interface.py
+++ fairseq/models/roberta/hub_interface.py
@@ -136,7 +136,7 @@
         doc.user_token_hooks['vector'] = lambda token: aligned_feats[token.i]
         return doc
 
-    def fill_mask(self, masked_input: str, topk: int = 5):
+    def fill_mask(self, masked_input: str, topk: int = 6):
         masked_token = ''
         assert masked_token in masked_input and masked_input.count(masked_token) == 1, \
             "Please add one {0} token for the input, eg: 'He is a {0} guy'".format(masked_token)